A R T ic L E s

Synapse maturation by activity-dependent ectodomain shedding of SIRPa

Anna B Toth1,4, Akiko Terauchi1,4, Lily Y Zhang1,4, Erin M Johnson-Venkatesh1,4, David J Larsen1, Michael A Sutton1,2 & Hisashi Umemori1,3
Formation of appropriate synaptic connections is critical for proper functioning of the brain. After initial synaptic differentiation, active synapses are stabilized by neural activity-dependent signals to establish functional synaptic connections. However, the molecular mechanisms underlying activity-dependent synapse maturation remain to be elucidated. Here we show that activitydependent ectodomain shedding of signal regulatory protein-a (SIRPa) mediates presynaptic maturation. Two target-derived molecules, fibroblast growth factor 22 and SIRPa, sequentially organize the glutamatergic presynaptic terminals during the initial synaptic differentiation and synapse maturation stages, respectively, in the mouse hippocampus. SIRP a drives presynaptic maturation in an activity-dependent fashion. Remarkably, neural activity cleaves the extracellular domain of SIRP a, and the shed ectodomain in turn promotes the maturation of the presynaptic terminal. This process involves calcium/calmodulin-dependent protein kinase, matrix metalloproteinases and the presynaptic receptor CD47. Finally, SIRP a-dependent synapse maturation has an impact on synaptic function and plasticity. Thus, ectodomain shedding of SIRP a is an activity-dependent trans-synaptic mechanism for the maturation of functional synapses. Synapses are the sites of information processing between neurons in the brain. Defects in synaptic circuitry in the hippocampus, a structure critical for long-term memory formation, emotional processing and social behavior, are associated with a variety of neurological and psychiatric disorders, including fragile X syndrome, autism, epilepsy and schizophrenia13. Thus, proper assembly of hippocampal synapses is essential for optimal functioning of the brain. To organize synapse formation, signals are exchanged between pre- and postsynaptic neurons. Two forms of signal are required for functional synapse formation during development: activity-independent and activity-dependent signals. Usually, initial synaptic differentiation is regarded as consisting of activity-independent steps, whereas a period of activity-dependent synapse maturation shapes the ultimate structure of neural circuits47. During synapse maturation, activity-dependent signals either stabilize or eliminate axons and further maturate selected synapses to establish appropriate synaptic connections812. Thus, activity-dependent mechanisms are required for the structural refinement of neural circuits, to match pre- and postsynaptic function, and for the final arrangement of the appropriate synaptic map1114. While synapse stabilization and destabilization and synapse maturation are clearly activity-dependent, little is known about molecular mechanisms underlying them. Defects in activity-dependent synapse maturation in the hippocampus have been implicated in various neurodevelopmental disorders, including schizophrenia and autism13. Therefore, an understanding of the molecules and manner by which hippocampal circuits are established by neural activity should yield new insights into both the etiology and treatment of these devastating disorders.
1Molecular

2013 Nature America, Inc. All rights reserved.

To understand the molecular mechanisms of synapse formation, we have performed an unbiased search for molecules that promote differentiation of axons into presynaptic nerve terminals. Using the ability to cluster synaptic vesicles in cultured motor neurons as a bioassay, we have purified from developing brains molecules that can promote differentiation of axons into presynaptic nerve terminals and identified two molecules, fibroblast growth factor 22 (FGF22)15 and SIRP16, as such presynaptic organizers. We have shown that FGF22 and its close relative FGF7 are selectively involved in the initial organization of excitatory (glutamatergic) and inhibitory (GABAergic) synapses, respectively, in the hippocampus17. The other molecule, SIRP, is a transmembrane immunoglobulin superfamily member that is involved in various hematopoietic cell functions1820, but little is known about its roles in the brain. We therefore investigated the role, mechanism and impact of SIRP-dependent synapse formation in the brain. Here we show that (i) target-derived FGF22 and SIRP sequentially organize presynaptic terminals; (ii) SIRP is necessary for presynaptic maturation, but not for induction or maintenance, in the hippocampus in vivo; (iii) SIRP drives presynaptic maturation in an activity-dependent manner; (iv) activity cleaves the ectodomain of SIRP, and this cleavage is required for SIRPs presynaptic effects; (v) calcium, calcium/calmodulin-dependent protein kinase (CaMK) and matrix metalloproteinase (MMP) mediate SIRP cleavage; (vi) CD47 is SIRPs presynaptic receptor; and (vii) SIRP has an impact on synaptic function and plasticity. These results indicate that ectodomain shedding of SIRP is an activity-dependent mechanism

& Behavioral Neuroscience Institute, University of Michigan Medical School, Ann Arbor, Michigan, USA. 2Department of Molecular and Integrative Physiology, University of Michigan Medical School, Ann Arbor, Michigan, USA. 3Department of Biological Chemistry, University of Michigan Medical School, Ann Arbor, Michigan, USA. 4These authors contributed equally to this work. Correspondence should be addressed to H.U. (umemoh@umich.edu or hisashi.umemori@childrens.harvard.edu). Received 15 May; accepted 13 August; published online 15 September 2013; doi:10.1038/nn.3516

NATURE NEUROSCIENCE

ADVANCE ONLINE PUBLICATION

A R T ic L E s
allowing pre- and postsynaptic terminals to communicate for the maturation of functional synapses. RESULTS Distinct expression of FGF and SIRPa during synaptogenesis We first compared the expression patterns of SIRP and FGFs in the hippocampus during synapse formation. In situ hybridization experiments with mouse brain sections showed little Sirpa mRNA expression in hippocampal neurons at postnatal day 8 (P8; Fig. 1a), an early stage of synapse formation21,22, but substantially higher expression at P21, a late stage of synapse formation. Western blotting confirmed a robust increase in the amount of SIRP protein from P8 to P21 (Fig.1b). This expression pattern is in contrast to the patterns of Fgf22 and Fgf7 mRNA, which were found to be highly expressed at P8 (ref. 17) but not at P21 (Supplementary Fig. 1). These results suggest that FGFs and SIRP are involved in the early and late stages of synapse formation, respectively, in the hippocampus. We next examined the localization of SIRP in hippocampal neurons. Biochemical fractionation experiments revealed that SIRP was abundant in the synaptic membrane fraction, indicating that SIRP is a synaptic molecule. Notably, it was most enriched in the extrajunctional fraction (Supplementary Fig. 2), which is similar to the distribution of some of other synaptogenic molecules, including EphB2 (ref. 23). Immunostaining of cultured hippocampal neurons showed that SIRP was preferentially localized at MAP2-positive dendrites relative to neurofilament-positive axons (Fig. 1c). In dendrites, SIRP was concentrated at excitatory synapses: it was colocalized (~75%) with vesicular glutamate transporter 1 (VGLUT1), a marker for glutamatergic presynaptic terminals, but showed little colocalization (~13%) with vesicular GABA transporter (VGAT), a marker for GABAergic presynaptic terminals (Fig. 1d). These results suggest that SIRP is localized in dendrites at glutamatergic synapses (that is, it is postsynaptic) and may serve as a target-derived glutamatergic presynaptic organizer in the hippocampus.
kDa 100 75 50
SIRP Intensity (% P0) 1,600

2013 Nature America, Inc. All rights reserved.

a
P8 CA3 CA1 DG

Sirpa mRNA P21

b
SIRP

P0

P8

P21

c d

SIRP/MAP2

SIRP/NF

-tubulin

800

SIRP/VGLUT1

SIRP/VGAT

e
GFP/Synapsin Control SIRP NLGN1
2

Control Synapsin

SIRP

NLGN1

f
VGLUT1
2

P0 P8 P2 1

Control VGLUT1

sSIRP

g
Frequency (Hz)

Puncta per mm (103)

Size (m )

Control 0.5 sSIRP 0

Size (m2)

1.0

Puncta per mm (103)

100 50 0

* ** *
2

* * * * **

**
0.5

**

mEPSCs Control sSIRP

Control 0.2

sSIRP

20 Amplitude (pA)

60 30 0

10

8 pA

nc

ue

eq

DIV0 1

11

i
DIV14 Control FGF22 sSIRP DIV48 Control FGF22 sSIRP DIV811 Control FGF22 sSIRP

j FGF22
Number (% control) 200 100 0

Size (% control)

Size (% control)

14

Number (% control)

VGLUT1

48 811 Treatment

*** * * * * *

*** * * *
100

sSIRP 200

*** * * * **
100

Fr

Am

pl

itu

de

500 ms

** * **

* * *

100

0
D IV D 1 I 4 D V4 IV 8 8 11

0
D IV D 1 D IV4 4 IV 8 8 11

0
D IV D 1 I D V4 4 IV 8 8 11

Figure 1 FGF22 and SIRP promote the early or late stage of glutamatergic WT presynaptic differentiation, respectively. *** Fgf22/ VGLUT1 ** Control + sSIRP + FGF22 ** (a) In situ hybridization for Sirpa in the * WT WT WT hippocampus (positive signals in black). 1.0 * 400 * Reproduced three times. (b) Western * * * * * * blotting for SIRP protein (-tubulin as * Fgf22/ Fgf22/ Fgf22/ * 0.5 * 200 control) in the hippocampus. Full-length blots are presented in Supplementary 0 0 Figure 10. (c,d) Hippocampal cultures (DIV11) stained with the antibodies indicated. ( c) SIRP protein were abundant on MAP2+ dendrites but not on neurofilament+ (NF+) axons. (d) SIRP was concentrated at VGLUT1+ glutamatergic synapses but not at VGAT+ GABAergic synapses. Reproduced five times. (e) HEK cells expressing SIRP, neuroligin1 (NLGN1) or control HEK cells (labeled with GFP) were cocultured with hippocampal neurons for 2 d and stained for synapsin. The synapsin puncta formed on HEK cells expressing SIRP were significantly more dense and larger than those formed on control HEK cells and were comparable to the ones on HEK cells expressing NLGN1. n = 24, 23 and 22 fields from 5 cultures. (f,g) sSIRP was applied to hippocampal cultures from DIV111. (f) sSIRP treatment significantly increased the number and size of VGLUT1 puncta as compared to PBS control ( n = 57 fields from 5 cultures). (g) Representative traces and summary data of whole-cell recordings of mEPSCs from control and sSIRP -treated hippocampal neurons. n = 57 and 63 cells from 5 cultures. (h) Schematic timeline of the experiment shown in i,j. Cultured hippocampal cells were treated with FGF22 or sSIRP from DIV14 (beginning of synaptogenesis), DIV48, or DIV811 (ending of synaptogenesis). All cultures were fixed on DIV11. (i) Staining of hippocampal cultures for VGLUT1. (j) Numbers and sizes of VGLUT1 puncta. Data are shown as percentage of PBS control. n = 32, 43, 40, 34, 27 and 26 fields from 5 cultures. (k) sSIRP or FGF22 was applied to hippocampal cultures prepared from WT or Fgf22/ mice, and the cultures were stained at DIV11. The number and size of VGLUT1 puncta on Py + (CA3 pyramidal neuron) dendrites were quantified. n = 19, 23, 25, 25, 17 and 23 neurites from 3cultures. Data are mean s.e.m. *P < 0.05, **P < 0.01, ***P < 0.0001 by Students t-test (f,g) or by ANOVA followed by Tukey test (e,j,k). Scale bars, 500 m (a), 10 m (e,k) and 5 m (others).

VGLUT1/PY

D IV D 1 I 4 D V4 IV 8 8 11

Number per mm

Size (m )

F2 2

IR

on

FG

on

IR

sS

sS

ADVANCE ONLINE PUBLICATION NATURE NEUROSCIENCE

FG

F2 2

tro

tro

A R T ic L E s
SIRPa promotes glutamatergic presynaptic differentiation To address whether SIRP can promote presynaptic differentiation of hippocampal neurons, we examined the effect that SIRP has on synaptic vesicle clustering using a coculture system24, in which neurons are cocultured with human embryonic kidney (HEK) cells. The number and size of synapsin puncta formed on HEK cells expressing SIRP were significantly larger than those on control HEK cells (Fig.1e). SIRPs presynaptic effects were comparable to those of neuroligin1, a well-characterized synaptogenic molecule, indicating that SIRP is a synaptogenic molecule that can promote synaptic vesicle clustering in hippocampal neurons. We then examined whether SIRP can organize glutamatergic presynaptic differentiation. For this, we prepared the extracellular portion of SIRP16 (soluble SIRP; sSIRP) and bath-applied it at 2 nM to cultured hippocampal neurons for 10 d. sSIRP significantly increased the number and size of VGLUT1 puncta (Fig.1f). Furthermore, SIRP increased the number and size of bassoon puncta, suggesting that SIRP organizes active zones as well (Supplementary Fig. 3a). Electrophysiological recordings indicated that sSIRP increased the frequency, but not the amplitude, of miniature excitatory postsynaptic currents (mEPSCs; Fig. 1g), consistent with an increase in synaptic contacts. sSIRP did not noticeably affect dendrite and axon differentiation or the clustering of PSD95, a postsynaptic scaffolding protein at glutamatergic synapses (Supplementary Fig. 3b,c), but it did increase the colocalization between VGLUT1 and PSD95 (Supplementary Fig. 3c). These results indicate that SIRP can specifically promote presynaptic differentiation of glutamatergic synapses in hippocampal neurons. FGF22 and SIRPa promote distinct stages of synaptogenesis On the basis of the developmentally different expression of Fgf22 and Sirpa mRNAs in the hippocampus, we hypothesized that FGF22 and SIRP are preferentially involved in the early and late stages of glutamatergic synapse formation, respectively. To test this idea, we performed time course experiments using cultured hippocampal neurons. In our hippocampal cultures, glutamatergic synapse formation starts around day in vitro (DIV) 3 and slows around DIV12 (ref.17). To determine the time during which FGF22 and SIRP are most effective at promoting presynaptic differentiation, we cultured hippocampal neurons and applied recombinant FGF22 or sSIRP during three different periods: DIV1DIV4, DIV4DIV8 or DIV8DIV11, which correspond to the beginning, middle and ending of synapse formation, respectively (Fig. 1h). We then stained all cultures for VGLUT1 at DIV11. FGF22 treatment was most effective at increasing the number and size of VGLUT1 puncta when applied from DIV1 DIV4 (Fig. 1i,j and Supplementary Fig. 4), consistent with an early role in synapse development. In contrast, sSIRP treatment increased the number and size of VGLUT1 puncta most prominently when it was applied from DIV8DIV11. These results support the notion that FGF22 and SIRP are presynaptic organizing molecules with temporally distinct roles during synapse formation, with FGF22 involved in early and SIRP in late stages of synapse formation. To further show that FGF22 and SIRP have distinct roles in presynaptic differentiation, we cultured FGF22-deficient neurons17 and examined whether SIRP could rescue their synaptic defects. There were fewer and smaller VGLUT1 puncta on CA3 neurons in FGF22deficient cultures relative to wild-type cultures. These presynaptic defects were rescued by the application of FGF22, but not by sSIRP, to cultures (Fig. 1k). These results suggest that, although both FGF22 and SIRP can induce presynaptic differentiation, their specific roles in presynaptic differentiation are different.
NATURE NEUROSCIENCE ADVANCE ONLINE PUBLICATION

SIRPa is required for presynaptic maturation in vivo In Fgf22/ mice, the differentiation of glutamatergic nerve terminals in the hippocampus is impaired early in synapse development, at P8 (ref. 17). To identify the developmental stages during which SIRP is critical for synapse formation in vivo, we generated a conditional SIRP knockout mouse (Supplementary Fig. 5a). To temporally control the expression of SIRP, we mated mice with loxP-flanked Sirpa alleles to actin-Cre-ER (CAG-cre/Esr1*) mice25 and injected their offspring with tamoxifen at different postnatal days to induce Cre-mediated excision of the Sirpa gene. Tamoxifen injections inactivated SIRP effectively in the hippocampus of these mice as confirmed by immunostaining for SIRP (Supplementary Fig. 5b). In the rodent hippocampus, synapse formation starts in the first postnatal week21,22. After their initial formation, synapses are then refined in an activity-dependent manner: effective synapses are stabilized and mature, while inactive contacts are destabilized and eliminated410. We have previously shown that activity-dependent synapse refinement in the hippocampus occurs between P15 and P25 (ref. 26). Thus, we chose three time periods corresponding to three different stages of synapse development to inactivate SIRP: P0P14 (initial synapse differentiation), P15P29 (synapse maturation) and P30P44 (synapse maintenance). When we injected SIRP conditional knockout mice with tamoxifen to generate Sirpa/ mice at P0 and stained their hippocampi for VGLUT1 at P14, the intensity of VGLUT1 staining was not significantly different (P = 0.514, stratum lucidum; 0.828, stratum radiatum) in the Sirpa/mice as compared to the wild-type littermate control (Fig. 2a), indicating that, unlike FGF22 (ref. 17), SIRP is not critical for initial synapse development. In contrast, when we injected tamoxifen at P15 and analyzed at P29, the intensity of VGLUT1 staining was significantly reduced in the knockout hippocampus as compared to controls (Fig. 2b). Further analyses revealed that the size and intensity of each VGLUT1 punctum were decreased in the Sirpa/ mice. In addition to VGLUT1, the intensity of bassoon staining, a marker for active zones, was also decreased in the Sirpa/ mice relative to that in control mice (Fig.2c). These results suggest that SIRP inactivation affects the maturation of presynaptic terminals. Finally, when we injected tamoxifen at P30 and analyzed at P44, the intensity of VGLUT1 staining was not significantly different in the knockout mice as compared to control (P = 0.556, stratum lucidum; 0.469, stratum radiatum; Fig. 2d). These results demonstrate that SIRP is critical for presynaptic maturation (P15P29) but is not necessary for initial synapse development or synapse maintenance in vivo. We also performed a series of experiments to examine whether the inactivation of SIRP primarily affects presynaptic maturation. In the Sirpa/ mice with tamoxifen injected at P15, the hippocampus looked anatomically normal, and the fate of the cells in the hippocampus appeared to be unchanged (Supplementary Fig. 6ad). In addition, the clustering of PSD95 was not significantly decreased in Sirpa/ as compared to control mice (P = 0.095, stratum lucidum; 0.134, stratum radiatum; Supplementary Fig. 6e). Thus, SIRP appears to be primarily involved in presynaptic maturation in the hippocampus invivo. Consistent with the Sirpa expression pattern (Fig. 1a and data not shown), presynaptic maturation in Sirpa/ mice was impaired throughout the hippocampus (Supplementary Fig. 6f), as well as in the cerebellum. The presynaptic defects in Sirpa/ mice were still present at P130 (Supplementary Fig. 6g), suggesting that SIRP inactivation prevents presynaptic maturation rather than just delaying it. To further confirm the role of SIRP in presynaptic maturation, we examined the ultrastructure of excitatory (asymmetric) synapses formed in the hippocampus in P29 Sirpa/ mice injected with
3

2013 Nature America, Inc. All rights reserved.

A R T ic L E s
Figure 2 SIRP is required for the maturation, but not induction or maintenance, of excitatory presynaptic terminals in the hippocampus invivo. (ad) SIRP was inactivated for the period from P0P14 (a), P15P29 (b,c) or P30P44 (d) by tamoxifen injections into conditional SIRP knockout mice. Control Cre +Sirpa +/+ littermates also received tamoxifen. Hippocampal sections were stained for VGLUT1 (a,b,d) or bassoon (c). Images are from CA3; positive signals in white. SL, stratum lucidum; SR, stratum radiatum. Only for mice injected with tamoxifen at P15 and analyzed at P29 did the VGLUT1 (b) and bassoon (c) staining intensities (in arbitrary units) significantly decrease in Sirpa/ mice relative to control. Quantifications of the size and intensity of VGLUT1 puncta are also shown in b. n = 12 and 20 sections from 3 and 5 mice. (e) Electron microscopic analysis of asymmetric (excitatory) synapses in CA3 of P29 mice injected with tamoxifen at P15. Right, high magnification views of synaptic vesicles (SVs). The numbers of SVs within 400 nm of the active zone and the numbers of docked vesicles are quantified. n = 30 synapses from 3 mice. *P < 0.05, **P < 0.01, ***P < 0.0001 by Students t-test (ae). Scale bars, 50 m (ad) and 100 nm (e). (f) Evoked fEPSPs recorded in acute slices from CA1 of Sirpa/ mice and control littermates (tamoxifen injections at P15, analyses at P29). Sample traces of fEPSP recordings are shown. In inputoutput curves (right), fEPSP slope, but not fiber volley amplitude, was significantly lower in Sirpa/ mice than in control littermates (P < 0.001 by two-way ANOVA; n = 9 and 13cells from 4mice). (g) Paired-pulse facilitation across a range of inter-stimulus intervals for evoked EPSCs from Sirpa/ mice and control littermates. Paired-pulse facilitation was significantly enhanced in Sirpa/ mice (P < 0.001 by two-way ANOVA; n = 11 and 15 cells from 4 mice). Data are mean s.e.m.

a
VGLUT1

P0 (tamoxifen injection) Control SL SR Sirpa/ SL SR

P14 (analysis) Average intensity Control Sirpa/ 1,000

P15 (tamoxifen injection) Control Sirpa/

P29 (analysis) Control

Sirpa/

SL

SR Number of SVs Control Sirpa/ 40

b
VGLUT1

P15 (tamoxifen injection) Control SL SR Sirpa/ SL SR

Average intensity

P29 (analysis) Control Sirpa/ 1,000

Synapse count

***

15 10 5 0 0 1 2 34 5 Number of docked vesicles

**
0

f
SL SR

***
Punctum intensity Punctum size (m2) 300 150 0 5

***

fEPSP Control
0.25 mV

Fiber volley amplitude 0.6 mV Control Sirpa/

10 ms

0.3

Sirpa/ 0 SL P29 (analysis) Average intensity 1,000 50 100 150 200 Stimulus intensity (A)

2013 Nature America, Inc. All rights reserved.

SL

c
Bassoon

P15 (tamoxifen injection) Control SL Sirpa/ SL

**
mV ms1 Control Sirpa/

EPSP slope 0.6 Control Sirpa/

500

0.3

SR

SR

0 0
SL

d
VGLUT1

P30 (tamoxifen injection) Control SL SR Sirpa/ SL Average intensity SR

P44 (analysis) Control Sirpa/ 1,000

50 100 150 200 Stimulus intensity (A) Paired-pulse facilitation Control 1.8
0.5 mV 20 ms

Control Sirpa/

Ratio

Sirpa/ 0

1.4

1.0 SL SR 50 100 150 200 Inter-stimulus interval (ms)

tamoxifen at P15 (Fig. 2e). We found significantly fewer synaptic vesicles and fewer docked synaptic vesicles in the asymmetric synapse in Sirpa/ mice relative to the number in controls. In addition, the shape of synaptic vesicles in Sirpa/ mice looked irregular compared to those in controls. Diminished transmitter release probability in Sirpa/ mice To directly address the functional state of synapses in Sirpa/ mice, we recorded evoked field excitatory postsynaptic potentials (fEPSPs) at CA3-CA1 synapses in acute hippocampal slices (tamoxifen injections at P15, analyses at ~P29). Input-output curves of fEPSP slope were strongly diminished in Sirpa/ mice relative to those in control littermates (Fig. 2f), whereas fiber volley amplitude (reflecting the number of axons firing to each stimulation) was unaffected, indicating that synaptic transmission is impaired in the absence of SIRP. Moreover, paired-pulse facilitation was dramatically increased in Sirpa/ mice relative to controls (Fig. 2g), suggesting that neurotransmitter release probability is diminished in the knockout mice. Sirpa/ neurons therefore have substantial defects in excitatory presynaptic function. Taken together, the histological and electrophysiological results from Sirpa/ mice demonstrate that SIRP is necessary for the maturation, but not induction or maintenance, of excitatory presynaptic terminals in the hippocampus in vivo.
4

Presynaptic maturation by SIRPa requires neural activity During the maturation stage of synapse formation, activity-dependent signals either stabilize or destabilize the synapses to establish efficient synaptic connections. Therefore, we hypothesized that SIRP contributes to mechanisms that stabilize and promote maturation of presynaptic terminals in response to neural activity. To test this idea, we examined whether the presynaptic effects of SIRP require neural activity. When SIRP was transfected into cultured hippocampal neurons, the VGLUT1 puncta were significantly larger on the dendrites of SIRP-expressing neurons than in controls (Fig. 3a,b). This SIRP-dependent increase in VGLUT1 clustering was completely blocked by suppressing neural activity with the sodium channel blocker tetrodotoxin (TTX, 1 M) or by suppressing synaptic transmission with a cocktail of neurotransmitter receptor inhibitors (50 M d-()-2-amino-5-phosphonopentanoic acid (AP5), 10 M 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX), 50 M bicuculline). These data suggest that neural activity is critical for SIRP-dependent presynaptic maturation. Neural activity cleaves the extracellular domain of SIRP a What are the mechanisms by which neural activity controls SIRPdependent presynaptic maturation? A clue came from our initial identification of SIRP as a presynaptic organizing molecule: the
ADVANCE ONLINE PUBLICATION NATURE NEUROSCIENCE

A R T ic L E s a
VGLUT1/GFP

b
Control Control + TTX Control + Inh Control Control + TTX Control + Inh SIRP SIRP + TTX SIRP + Inh

c
*
C el ll ys at e M ed iu m

d
C KCl sSIRP kDa (medium) 100 -tubulin (lysate) 75 kDa 75

50

SIRP

SIRP + TTX

SIRP + Inh

Size (m2)

0.5

f
Percent control 300 200 100 0
C

sSIRP Percent control

* * *

Full-length SIRP

g
Full-length SIRP (lysate) C KCl TTX kDa 100 75

WT

MT

* *
100 0
C on tro KCl TT l X

sSIRP (medium)

kDa 75 100

e sSIRP

Bic

TTX

kDa 75

(medium) -tubulin (lysate) 50

* *

* *

Full-length SIRP (lysate)

on tro KCl TT l X

i
GFP/Synapsin Control WT-SIRP MT-SIRP

VGLUT1/GFP Control

k
VGLUT1
Control sSIRP # TTX TTX + sSIRP

l
GFP/VGLUT1 Control

TTX Control Ext-SIRP

+ TTX

WT-SIRP
Control WT-SIRP MT-SIRP

Ext-SIRP

2013 Nature America, Inc. All rights reserved.

Synapsin Puncta per mm2 (103) 100 50 0

MT-SIRP

Size (m2)

1.0 0.5 0

Size (m2)

Size (m )

*
0.5

300 150 0

0.5

Size (m2)

* *

* * *

Control WT-SIRP MT-SIRP

Control TTX sSIRP TTX + sSIRP

Number per mm

* * * * *

VGLUT1

*** ***

** **

TTX + TTX

1.0 0.5 0

SI Ext R P

Figure 3 Cleavage of the extracellular domain of SIRP is activity-dependent and is necessary for SIRP -dependent presynaptic maturation. (a,b) Hippocampal neurons were transfected with a GFP plasmid (Control) or SIRP and GFP plasmids (SIRP) at DIV4. TTX or an inhibitor cocktail (Inh: AP5, CNQX, bicuculline) was applied from DIV5. Cultures were stained on DIV13. VGLUT1 clusters on the dendrites of SIRP -transfected neurons were larger than those of control neurons, but addition of TTX or Inh prevented this increase. n = 10 neurites from 3 cultures. (c) Hippocampal neurons were cultured for 11 d. Media and cell lysates were assayed for SIRP . Reproduced five times. (dg) Hippocampal neurons were cultured for 1012 d and then incubated with PBS (control; C), KCl, bicuculline (Bic) or TTX for 13 d. Medium was assayed for sSIRP (d,e); cell lysates were assayed for -tubulin (d,e) and full-length SIRP remaining on the cell (g). (d) Addition of KCl increased the amount of sSIRP in the medium as compared to control. (e) Bicuculline increased the amount of sSIRP , whereas TTX decreased it. (f) Quantification of the band intensity from results such as those shown in d,e,g. Intensities were normalized against the intensity of the control band. n = 5 and 3 blots from 5 and 3 cultures. (g) In the cell lysate, the amount of full-length SIRP was decreased with KCl and increased with TTX. ( h) Verification of cleavage-resistant mutant (MT)-SIRP . MT-SIRP showed no release of sSIRP from transfected HEK cells. Reproduced three times. ( i) HEK cells expressing WT- or MT-SIRP and control HEK cells (labeled with GFP) were cocultured with hippocampal neurons for 2 d and stained for synapsin. MT-SIRP did not increase the number and size of synapsin puncta. n = 24, 24 and 33 fields from 5 cultures. (j) Hippocampal neurons were transfected with the GFP plasmid (Control) or the WT- or MT-SIRP plasmid together with the GFP plasmid at DIV4 and stained at DIV13. VGLUT1 puncta on the dendrites of WT-SIRP transfected neurons, but not on those of MT-SIRPtransfected neurons, were larger than control. n = 23 neurites from 3 cultures. (k) Hippocampal neurons were treated with sSIRP and/or TTX from DIV1 and stained at DIV13. sSIRP increased the size of VGLUT1 puncta, and adding TTX had no effect on this increase. n = 13 fields from 3 cultures. (l) Hippocampal neurons were transfected with the GFP plasmid (Control) or secretable (Ext)-SIRP and GFP plasmids at DIV4. TTX was applied from DIV5. Cultures were stained on DIV13. VGLUT1 clusters on the dendrites of Ext-SIRP transfected neurons were greater in number and size than those of control neurons, and addition of TTX had no effect on this increase. n = 26, 23, 21 and 24 neurites from 3 cultures. Data are mean s.e.m. *P < 0.05, **P < 0.01, ***P < 0.0001 by Students t-test (b,f) or ANOVA followed by Tukey test (il). #, not significant (P = 0.998, 0.989 (i); 0.964 (k)). Scale bars, 10 m (i,l), 5 m (a,j,k).

SIRP protein we identified from the brain extract was the extracellular portion of SIRP16. In fact, from cultured neurons, we were able to collect secreted SIRP in the media, and its molecular weight was smaller than full-length SIRP expressed in neurons (Fig. 3c). In addition, we detected a short fragment of SIRP containing its C-terminal domain (~16 kDa, which corresponds to the intracellular domain) in the synaptic membrane fraction (Supplementary Fig. 2). Therefore, we hypothesized that the extracellular domain of SIRP is cleaved by neural activity and that this cleavage is required for its presynaptic effects (see Supplementary Fig. 7a). To examine whether the extracellular domain of SIRP is cleaved and released from hippocampal neurons in response to neural activity, we cultured hippocampal cells with either KCl (50 mM) to depolarize neurons, bicuculline (50 M) to enhance endogenous network
NATURE NEUROSCIENCE ADVANCE ONLINE PUBLICATION

activity or TTX (1 M) to suppress network activity. We then collected media and assessed the amount of cleaved and released SIRP by immunoprecipitation followed by western blot. KCl and bicuculline treatments significantly increased the amount of released SIRP in media as compared to that in untreated control (Fig. 3df), while TTX treatment significantly decreased the amount of cleaved SIRP in the media, indicating that the SIRP ectodomain is released by neural activation. These effects were not due to altered cell numbers, as the amount of -tubulin in the cell lysate was not altered by any treatment condition. In addition, the amount of full-length SIRP remaining on the cell was decreased in KCl-treated cultures and increased in TTX-treated cultures (Fig.3f,g), consistent with an increase or a decrease in SIRP cleavage by KCl or TTX treatment, respectively.
5

SI Ext R P

tro

on

on

tro

A R T ic L E s
Percent control

a
sSIRP (medium) -tubulin (lysate) C KN62 kDa 75 50 sSIRP (medium) -tubulin (lysate)

KN93

kDa 75 50

100 50 0

* *

sSIRP (medium) -tubulin (lysate)

kDa 75 50

Percent control

** **
KCl KCl KN62 KCl Nif 400 300 200 100 0
KC K l K Cl N KC 62 lN if

d
sSIRP (medium) -tubulin (lysate) C TIMPs kDa 75 50

Figure 4 SIRP-dependent presynaptic maturation involves calcium channels, CaMK and MMP. (ad) Hippocampal neurons were cultured for 1012 d and then incubated with a calcium channel blocker (nifedipine; Nif), CaMK inhibitors (KN62 or KN93) or MMP inhibitors (GM6001 (GM) or TIMPs), in the presence (b,c) or absence of KCl for 13 d. C, control (PBS or DMSO). Media were collected and assayed for the amount of sSIRP. Cell lysates were assayed for -tubulin. Quantification of the sSIRP band intensity is shown in the graphs (normalized against control). Nifedipine, KN62, KN93, GM6001 and TIMPs decreased the amount of sSIRP found in the media. n = 4 blots from 4 cultures. Data are mean s.e.m. *P < 0.05, **P < 0.01 by Students t-test.

tr KN ol 62

Percent control

C on t KNrol 93

C on

* *
100 50 0

C sSIRP (medium) -tubulin (lysate)

GM

KCl KCl GM

Percent control

kDa 75

300 200 100

50

0 KCl GM6001 +

+ + +

size of VGLUT1 puncta on the transfected neurons; however, overexpression of MT-SIRP failed to do so (Fig. 3j). These results indicate that shedding-resistant SIRP cannot promote presynaptic maturation both in coculture and neuronal culture, suggesting that the cleavage and secretion of the SIRP ectodomain are necessary for its presynaptic effects. Neural activity is responsible for SIRPa cleavage If neural activity is responsible for cleaving SIRP, suppressing neural activity should inhibit the presynaptic effect of full-length SIRP (Fig. 3a,b) but not that of soluble SIRP (sSIRP). To test this idea, we cultured hippocampal neurons with sSIRP with or without TTX. Application of sSIRP increased the size of VGLUT1 puncta, and, unlike what was observed with full-length SIRP, this effect was

2013 Nature America, Inc. All rights reserved.

Shedding of SIRPa is necessary for presynaptic maturation We then investigated whether the cleavage of the extracellular domain of SIRP is required for presynaptic maturation mediated by SIRP. For this, we prepared a mutant form of SIRP (MT-SIRP) that is resistant to ectodomain shedding (Fig. 3h). In the HEK cell hippocampal neuron coculture system (see Fig. 1e), the number and size of synapsin puncta formed on HEK cells expressing MT-SIRP were similar to those on control HEK cells (Fig. 3i), indicating that MT-SIRP cannot promote synaptic vesicle clustering in hippo campal neurons. We next transfected cultured hippocampal neurons with wildtype SIRP (WT-SIRP) or MT-SIRP. The localization of MT-SIRP was similar to that of WT-SIRP (Supplementary Fig.7b,c). Overexpression of WT-SIRP led to an increase in the
Figure 5 CD47 is the presynaptic receptor for SIRP-mediated presynaptic maturation. (a,b) In cultured hippocampal neurons, CD47 puncta were abundant in neurofilament+ axons and not in MAP2+ dendrites (a) and colocalized with SIRP (b). Reproduced five times. (c) Hippocampal neurons prepared from Cd47/ mice and WT littermates were treated with sSIRP and stained for VGLUT1. sSIRP did not increase the number and size of VGLUT1 puncta in Cd47/ neurons. n = 80, 93, 102 and 108 fields from 4 cultures. (d) HEK cells expressing SIRP, neuroligin1 or control HEK cells (labeled with GFP) were cocultured with hippocampal neurons from WT or Cd47/ mice for 2 d and stained for synapsin. HEK cells expressing SIRP induced presynaptic differentiation in cocultured WT hippocampal neurons, but failed to do so in cocultured Cd47/ neurons. HEK cells expressing neuroligin1 induced presynaptic differentiation in both WT and Cd47/ neurons. n = 38, 34, 34, 35, 42 and 40 fields from 5 cultures. (e) Hippocampal neurons from WT or Cd47/ mice were transfected with a synaptophysin (Syn)-YFP plasmid or with synaptophysinYFP and CD47 plasmids and then treated with sSIRP. sSIRP did not increase the number and size of synaptophysin-YFP puncta in Cd47/ neurons, but introduction of CD47 restored responsiveness to sSIRP. n = 33, 37, 33, 35, 37 and 36 neurites from 3 cultures. (f) Hippocampal neurons from SIRP conditional knockout (KO) mice were transfected with an mCherry plasmid (control), Cre and mCherry plasmids (knockout) or Cre, mCherry and SIRP plasmids (rescue) and stained for VGLUT1. Both the number and size of VGLUT1 puncta on mCherry-expressing dendrites were smaller in Sirpa/ than in control, but postsynaptic expression of SIRP rescued the defects. n = 41, 33 and 36 neurites from 3 cultures. Data are mean s.e.m. *P < 0.05, **P < 0.01, ***P < 0.0001 relative to WT (c,d) or control (e,f) by Students t-test (c) or ANOVA followed by Tukey test (df). #, not significant (P = 0.541, 0.997, 0.972, 0.285 (d); 0.981, 0.941 (e)). Scale bar, 5 m (c), 10 m (all others).

tr TI ol M Ps

C on

Neurofilament/CD47

c
VGLUT1
Control sSIRP

WT Control sSIRP WT

Cd47 /

MAP2/CD47

WT + sSIRP Cd47 / + sSIRP Size (% control)

Cd47 / SIRP/CD47 Number (% control) 100

* *

100

d
WT

GFP/Synapsin Control SIRP NLGN1

0 Synapsin Puncta per mm2 (103) 100 50 0 # WT

0 Cd47 /

Size (m2)

Control Cd47 /

SIRP

NLGN1

* * *
1 N

1.0 0.5 0

* * *

tro l P

l P R SI N
1.0 0.5 0

tro

LG

SI

e
Syn-YFP Control sSIRP WT Syn-YFP Number per mm 200 100

f
sSIRP Cd47 Control
/

sSIRP Cd47
/

VGLUT1/mCherry

Control

Control

Control Cre Cre + SIRP Control Cre Cre + SIRP Size (m2)

+ CD47

sSIRP

* * *
#

* * *
Size (m2) 1.0 0.5

* * *
#

* * *
Number per mm

VGLUT1 400 200 0

SIRP-KO

LG

on

on

0
/

C + d47 C D 47

d4 7

ADVANCE ONLINE PUBLICATION NATURE NEUROSCIENCE

C + d47 C D 47

d4 7

* * *

* * *

A R T ic L E s
completely resistant to TTX (Fig. 3k). Thus, after cleavage, presynaptic maturation by SIRP no longer depends on neural activity. To exclude the possibility that SIRP is subjected to constitutive cleavage in neurons followed by activity-dependent secretion of its cleaved product, we performed an experiment with a secretable form of SIRP that contains only the extracellular domain of SIRP (ExtSIRP; the construct used to prepare sSIRP). When transfected into cultured neurons, Ext-SIRP efficiently induced maturation of glutamatergic presynaptic terminals on the Ext-SIRPexpressing neurons (Fig. 3l). This presynaptic effect was not inhibited by TTX application, indicating that neural activity is not important for the secretion of Ext-SIRP. These results are consistent with the notion that neural activity is responsible for cleaving, and not secretion, of the extracellular domain of SIRP. CaMK and MMP mediate activity-dependent SIRPa cleavage We further investigated the signaling pathway that is involved in activity-dependent SIRP cleavage. CaMK is a major signaling molecule at synapses27, prompting us to explore the possibility that CaMK contributes to SIRP cleavage. Consistent with this hypothesis, treatment of hippocampal cultures with CaMK inhibitors, KN62 or KN93 (5 M), significantly decreased the amount of cleaved SIRP in the media (Fig. 4a). We next examined the effects of a CaMK inhibitor (KN62) and a calcium channel blocker (nifedipine; 10 M) on activity-dependent cleavage of SIRP. Both inhibitors suppressed KCl-induced SIRP cleavage (Fig. 4b), suggesting that neural activity-dependent calcium entry followed by CaMK activation are important for the ectodomain shedding of SIRP. We then characterized the proteases that cleave the extracellular domain of SIRP. MMPs are zinc-dependent endopeptidases that cleave extracellular molecules and are implicated in synaptic function28. We found that incubation of hippocampal neurons with MMP inhibitors, GM6001 (10 M) or tissue inhibitors of metalloproteinases (TIMPs) (0.5 g ml1), markedly inhibited SIRP shedding, including the augmented cleavage induced by KCl (Fig. 4c,d). These results suggest that calcium, CaMK and MMP are involved in the activity-dependent shedding of SIRP from hippocampal neurons, although it is known that any inhibitors have off-target effects, and we cannot completely rule out an influence of such effects at this time. CD47 is the presynaptic receptor for SIRPa Because our data indicated that the shed SIRP ectodomain promotes presynaptic maturation, we next examined the identity of its presynaptic receptor. We asked whether CD47, a receptor for SIRP in hematopoietic cells1820, mediates the presynaptic effects of SIRP. Immunostaining experiments showed that CD47 puncta were abundant in neurofilament-positive axons and not in MAP2-positve dendrites (Fig. 5a) and that CD47 colocalized with SIRP (Fig. 5b), consistent with the idea that CD47 serves as a presynaptic receptor for SIRP. We then used Cd47/ neurons29 to determine whether CD47 mediates SIRPs effects. Cd47/ neurons extended axons and dendrites normally (Supplementary Fig. 8) but did not increase the number and size of VGLUT1 puncta in response to sSIRP application (Fig. 5c). The following two experiments suggest that CD47 acts as a presynaptic receptor for SIRP: (i) HEK cells expressing SIRP, which can induce presynaptic differentiation in cocultured WT hippocampal neurons, failed to do so in cocultured Cd47/ neurons, while neuroligin1 was able to induce presynaptic differentiation in both WT and Cd47/ neurons (Fig. 5d), and (ii) Cd47/ neurons did not respond to sSIRP application to induce presynaptic differentiation
NATURE NEUROSCIENCE ADVANCE ONLINE PUBLICATION
LTP Control Sirpa/
0.25 mV

10 ms

220
Normalized fEPSP slope (%)

Control Sirpa/

160

100 0 50 150 100 Time (min) 200

2013 Nature America, Inc. All rights reserved.

Figure 6 Impact of SIRP-dependent presynaptic maturation on synaptic plasticity. Hippocampal slices were prepared from Sirpa/ mice and control littermates and fEPSPs recorded at CA3CA1 synapses. LTP was induced by high-frequency stimulation (four trains of 1-s, 100-Hz stimulations spaced at 30-s intervals). Green is before and blue is 1 h after the LTP induction. LTP was significantly impaired in Sirpa/ mice (P < 0.05 by Students t-test at 1 h after the LTP induction). n = 10 and 20 slices from 7 and 11 mice. Data are mean s.e.m.

as assessed by synaptophysin-YFP clustering, but the responsiveness was restored by presynaptic expression of CD47 (Fig. 5e). Finally, we confirmed that the source of SIRP is postsynaptic: we found that presynaptic defects in Sirpa/ neurons were rescued by postsynaptic expression of SIRP (Fig. 5f). Altogether, these results strongly suggest that postsynaptically derived SIRP interacts with presynaptic CD47 to organize presynaptic maturation. LTP is impaired in Sirpa/ mice What are the functional consequences of defects in SIRP-dependent presynaptic maturation? To explore this question, we examined the impact of SIRP-deficiency on activity-dependent synaptic plasticity in the hippocampus. Long-term potentiation (LTP) was impaired at CA3CA1 synapses in the hippocampus of Sirpa/ mice (Fig. 6); of relevance, Cd47/ mice also show impaired LTP30. This is consistent with the altered presynaptic function in Sirpa/ mice (Fig. 2 and Supplementary Fig. 9) and demonstrates that SIRP-dependent synapse maturation has an enduring impact on long-lasting forms of plasticity in hippocampal circuits. DISCUSSION Activity-dependent synapse maturation is a critical step for the refinement of neural circuits and the establishment of an appropriate and efficient synaptic map in the brain. However, little is known about the molecular mechanisms that control this important aspect of synapse development. Here we have uncovered a new process by which neural activity contributes to synapse maturation. From our results, we propose that, after initial synaptic differentiation by molecules such as FGF22 (ref. 17), synaptic activity regulates extracellular domain cleavage of postsynaptic SIRP through CaMK and MMP, and the released SIRP ectodomain, in turn, promotes the maturation of the presynaptic terminal through CD47 (Supplementary Fig. 7a). SIRP-dependent synapse maturation affects synaptic function and plasticity, as demonstrated by impaired basal transmission, diminished neurotransmitter release probability and impaired LTP in Sirpa/ mice.
7

A R T ic L E s
Many molecules have been implicated in presynaptic development, including neuroligins, SynCAMs, ephrins and Eph receptors, leucinerich repeat transmembrane neuronal proteins (LRRTMs), netrin-G ligands (NGLs), FGFs, Wnts, neurotrophins, cerebellin and thrombospondins57,3133. Why are there multiple presynaptic organizers? We hypothesized that different presynaptic organizers exist for the organization of different types of synapses (spatial specificity) and for the regulation of different stages of synapse formation (temporal specificity). Concerning spatial specificity, we have previously shown that two FGFs, FGF22 and FGF7, are involved in the differentiation of two distinct types of synapses in vivo: FGF22 in excitatory and FGF7 in inhibitory17. Several presynaptic organizers, such as neuroligin1, SynCAMs, Eph receptors, LRRTMs and NGLs, seem to be relatively specific to excitatory synapses, whereas others, such as neuroligin2 and BDNF, may be preferentially involved in inhibitory synapses. Thus, distinct presynaptic organizers indeed appear to contribute to the organization of different synapses in the brain. As for temporal specificity, we have here shown that, in the hippocampus, FGF22 and SIRP are important for two sequential stages of synapse formation, with FGF22 influencing initial synaptic differentiation and SIRP influencing synapse maturation. Together, we propose that multiple spatially and temporally defined presynaptic organizers cooperate to organize specific and functional synaptic networks in the brain. The role of SIRP has been mainly studied in the immune system. Little is known about its function in the nervous system, but possible roles for SIRPs intracellular domain have been suggested: it promotes neurite outgrowth and enhances the effect of BDNF in culture20,34, and mice expressing mutant SIRP that lacks the intra cellular domain show prolonged immobility in the forced swim test 35. We focused on the role of SIRPs extracellular domain: using hippo campal cultures and conditional SIRP knockout mice, we found that the extracellular domain of SIRP serves as a target-derived presynaptic organizer in the hippocampus and is critical in the maturation stage of synapse formation in vitro and in vivo. SIRPs extracellular domain is cleaved in response to neural activity, acting as an activitydependent, target-derived presynaptic organizer. Why does SIRP need to be cleaved for presynaptic maturation? Cleavage may be necessary for the extracellular domain of SIRP to bind to its presynaptic receptor, CD47. Crystal structure models of SIRP and CD47 suggest that the extracellular region of the SIRP-CD47 complex is ~14 nm in length36,37. However, the cleft of excitatory synapses is ~25nm across, which may require the release of SIRP ectodomain to bind to CD47. It will be also interesting to address the fate and roles of the SIRP intracellular domain after cleavage. Ectodomain shedding is important in various processes, including sperm-egg interaction, cell migration and adhesion, cell fate determination, wound healing, axon guidance and immune responses3840. Here we have identified a new role for ectodomain shedding: activitydependent ectodomain shedding of SIRP is involved in synapse maturation. Notably, while we were preparing our paper, two groups showed that activity-dependent cleavage of neuroligin1 is involved in synapse disassembly and negatively regulates synaptic function in a homeostatic manner41,42. In contrast, our results demonstrate that activity-dependent cleavage of SIRP is a critical positive regulator of synapse maturation during synapse development for the establishment of functional circuits. It is also noteworthy that the cleavages of both SIRP and neuroligin1 involve common pathways, CaMK and MMP, yet they have opposite effects at synapses. Thus, our results, together with the neuroligin results, expand the role of activity-dependent shedding in controlling synapse maturation and function. How activity-dependent shedding of SIRP and neuroligin1
8

cooperate or antagonize to regulate synapses will be an interesting question to address next. Finally, since defects in activity-dependent synapse maturation in the hippocampus have been implicated in various neurological and psychiatric disorders, such as schizophrenia and autism13, our results may help design strategies to prevent and treat such disorders. Methods Methods and any associated references are available in the online version of the paper.
Note: Any Supplementary Information and Source Data files are available in the online version of the paper. ACKNOWLEDGMENTs We thank J. Sanes for critical comments on the manuscript; H. Enomoto for pSV loxP sv40 intron polyA EGFP FRTneo plasmid; A. Murayama and L. Kee for plasmid construction; E. Gibbs for help with in situ hybridization; D. Sorenson for help with electron microscopy; M. Zhang, R. Carson and A. Williams for technical assistance; and E. Hughes, Y. Qu, K. Childs, G. Gavrilina, D. Vanheyningen and the Transgenic Animal Model Core of the University of Michigan for preparation of SIRP knockout mice. Core support was provided by the University of Michigan Center for Organogenesis. This work was supported by the Ester A. & Joseph Klingenstein Fund, the Edward Mallinckrodt Jr. Foundation, the March of Dimes Foundation, the Whitehall Foundation and US National Institutes of Health grants MH091429, NS070005 and MH092614 (H.U.). AUTHOR CONTRIBUTIONS H.U. designed experiments and prepared the manuscript. A.B.T., A.T., L.Y.Z., E.M.J.-V. and D.J.L. performed experiments. M.A.S. and H.U. supervised the project. All authors analyzed data and commented on the manuscript. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.
Reprints and permissions information is available online at http://www.nature.com/ reprints/index.html.
1. Lipska, B.K., Halim, N.D., Segal, P.N. & Weinberger, D.R. Effects of reversible inactivation of the neonatal ventral hippocampus on behavior in the adult rat. J. Neurosci. 22, 28352842 (2002). 2. Pfeiffer, B.E. et al. Fragile X mental retardation protein is required for synapse elimination by the activity-dependent transcription factor MEF2. Neuron 66, 191197 (2010). 3. Kasai, H., Fukuda, M., Watanabe, S., Hayashi-Takagi, A. & Noguchi, J. Structural dynamics of dendritic spines in memory and cognition. Trends Neurosci. 33, 121129 (2010). 4. Sanes, J.R. & Lichtman, J.W. Development of the vertebrate neuromuscular junction. Annu. Rev. Neurosci. 22, 389442 (1999). 5. Waites, C.L., Craig, A.M. & Garner, C.C. Mechanisms of vertebrate synaptogenesis. Annu. Rev. Neurosci. 28, 251274 (2005). 6. Fox, M.A. & Umemori, H. Seeking long-term relationship: axon and target communicate to organize presynaptic differentiation. J. Neurochem. 97, 12151231 (2006). 7. Dalva, M.B., McClelland, A.C. & Kayser, M.S. Cell adhesion molecules: signaling functions at the synapse. Nat. Rev. Neurosci. 8, 206220 (2007). 8. Goda, Y. & Davis, G.W. Mechanisms of synapse assembly and disassembly. Neuron 40, 243264 (2003). 9. Tessier, C.R. & Broadie, K. Activity-dependent modulation of neural circuit synaptic connectivity. Front. Mol. Neurosci. 2, 8 (2009). 10. Kano, M. & Hashimoto, K. Synapse elimination in the central nervous system. Curr. Opin. Neurobiol. 19, 154161 (2009). 11. Zhang, L.I. & Poo, M. Electrical activity and development of neural circuits. Nat. Neurosci. 4, 12071214 (2001). 12. Bleckert, A. & Wong, R.O. Identifying roles for neurotransmission in circuit assembly: insights gained from multiple model systems and experimental approaches. Bioessays 33, 6172 (2011). 13. Kay, L., Humphreys, L., Eickholt, B.J. & Burrone, J. Neuronal activity drives matching of pre- and postsynaptic function during synapse maturation. Nat. Neurosci. 14, 688690 (2011). 14. Flavell, S.W. & Greenberg, M.E. Signaling mechanisms linking neuronal activity to gene expression and plasticity of the nervous system. Annu. Rev. Neurosci. 31, 563590 (2008). 15. Umemori, H., Linhoff, M.W., Onitz, D.M. & Sanes, J.R. FGF22 and its close relatives are presynaptic organizing molecules in the mammalian brain. Cell 118, 257270 (2004).

2013 Nature America, Inc. All rights reserved.

ADVANCE ONLINE PUBLICATION NATURE NEUROSCIENCE

A R T ic L E s
16. Umemori, H. & Sanes, J.R. Signal regulatory proteins (SIRPs) are secreted presynaptic organizing molecules. J. Biol. Chem. 283, 3405334061 (2008). 17. Terauchi, A. et al. Distinct FGFs promote differentiation of excitatory and inhibitory synapses. Nature 465, 783787 (2010). 18. van Beek, E.M., Cochrane, F., Barclay, A.N. & van den Berg, T.K. Signal regulatory proteins in the immune system. J. Immunol. 175, 77817787 (2005). 19. Barclay, A.N. & Brown, M.H. The SIRP family of receptors and immune regulation. Nat. Rev. Immunol. 6, 457464 (2006). 20. Matozaki, T., Murata, Y., Okazawa, H. & Ohnishi, H. Functions and molecular mechanisms of the CD47-SIRP signaling pathway. Trends Cell Biol. 19, 7280 (2009). 21. Danglot, L., Triller, A. & Marty, S. The development of hippocampal interneurons in rodents. Hippocampus 16, 10321060 (2006). 22. Steward, O. & Falk, P.M. Selective localization of polyribosomes beneath developing synapses: a quantitative analysis of the relationships between polyribosomes and developing synapses in the hippocampus and dentate gyrus. J. Comp. Neurol. 314, 545557 (1991). 23. Bouvier, D. et al. Pre-synaptic and post-synaptic localization of EphA4 and EphB2 in adult mouse forebrain. J. Neurochem. 106, 682695 (2008). 24. Biederer, T. & Scheiffele, P. Mixed-culture assays for analyzing neuronal synapse formation. Nat. Protoc. 2, 670676 (2007). 25. Guo, C., Yang, W. & Lobe, C.G. A Cre recombinase transgene with mosaic, widespread tamoxifen-inducible action. Genesis 32, 818 (2002). 26. Yasuda, M. et al. Multiple forms of activity-dependent competition rene hippocampal circuits in vivo. Neuron 70, 11281142 (2011). 27. Wayman, G.A., Lee, Y.S., Tokumitsu, H., Silva, A.J. & Soderling, T.R. Calmodulinkinases: modulators of neuronal development and plasticity. Neuron 59, 914931 (2008). 28. Ethell, I.M. & Ethell, D.W. Matrix metalloproteinases in brain development and remodeling: synaptic functions and targets. J. Neurosci. Res. 85, 28132823 (2007). 29. Lindberg, F.P. et al. Decreased resistance to bacterial infection and granulocyte defects in IAP-decient mice. Science 274, 795798 (1996). 30. Chang, H.P., Lindberg, F.P., Wang, H.L., Huang, A.M. & Lee, E.H. Impaired memory retention and decreased long-term potentiation in integrin-associated proteindecient mice. Learn. Mem. 6, 448457 (1999). 31. Shen, K. & Cowan, C.W. Guidance molecules in synapse formation and plasticity. Cold Spring Harb. Perspect. Biol. 2, a001842 (2010). 32. Williams, M.E., de Wit, J. & Ghosh, A. Molecular mechanisms of synaptic specicity in developing neural circuits. Neuron 68, 918 (2010). 33. Siddiqui, T.J. & Craig, A.M. Synaptic organizing complexes. Curr. Opin. Neurobiol. 21, 132143 (2011). 34. Wang, X.X. & Pfenninger, K.H. Functional analysis of SIRP in the growth cone. J. Cell Sci. 119, 172183 (2006). 35. Ohnishi, H. et al. Stress-evoked tyrosine phosphorylation of signal regulatory protein regulates behavioral immobility in the forced swim test. J. Neurosci. 30, 1047210483 (2010). 36. Hatherley, D. et al. Paired receptor specicity explained by structures of signal regulatory proteins alone and complexed with CD47. Mol. Cell 31, 266277 (2008). 37. Hatherley, D., Graham, S.C., Harlos, K., Stuart, D.I. & Barclay, A.N. Structure of signal-regulatory protein : a link to antigen receptor evolution. J. Biol. Chem. 284, 2661326619 (2009). 38. Edwards, D.R., Handsley, M.M. & Pennington, C.J. The ADAM metalloproteinases. Mol. Aspects Med. 29, 258289 (2008). 39. Reiss, K. & Saftig, P. The a disintegrin and metalloprotease (ADAM) family of sheddases: physiological and cellular functions. Semin. Cell Dev. Biol. 20, 126137 (2009). 40. Bai, G. & Pfaff, S.L. Protease regulation: the yin and yang of neural development and disease. Neuron 72, 921 (2011). 41. Peixoto, R.T. et al. Transsynaptic signaling by activity-dependent cleavage of neuroligin-1. Neuron 76, 396409 (2012). 42. Suzuki, K. et al. Activity-dependent proteolytic cleavage of neuroligin-1. Neuron 76, 410422 (2012).

2013 Nature America, Inc. All rights reserved.

NATURE NEUROSCIENCE

ADVANCE ONLINE PUBLICATION

ONLINE METHODS

In situ hybridization. In situ hybridization was performed as described43 using digoxigenin-labeled riboprobes (Roche). The probes were generated by PCR from the 3 untranslated regions1517. Primary neuronal cultures and transfection. Hippocampal cultures were prepared as described17. For immunostaining, hippocampal cells (1.5 104 to 4 104) were plated on glass coverslips (diameter 12 mm) coated with poly-dlysine. Transfection was performed using the CalPhos Mammalian transfection kit (Clontech). For immunoprecipitation, hippocampal cells (3 105 to 5 105) were plated on poly-d-lysinecoated tissue culture dishes (diameter 35 mm). For coculture experiments, HEK cells were transfected using Lipofectamine 2000 (Invitrogen), and 24 h after transfection they were dissociated and added onto cultured hippocampal neurons (DIV 8). Cocultures were maintained for 48 h before fixation. Knockout and transgenic mice. SIRP knockout mice: A Sirpa genomic clone containing exon 1 (BAC clone 394B7; Invitrogen) was used to construct a targeting vector. A gene cassette composed of floxed full-length mouse Sirpa cDNA with SV40 intron-poly(A), EGFP-poly(A) and FRTed Tn5 neo44 was introduced into the first exon, deleting 71 nucleotides containing the start codon (Supplementary Fig. 5). The deletion disrupts the expression of the endogenous Sirpa gene but allows the expression of the inserted gene. Floxed SIRP mice were generated by embryonic stem cellbased homologous recombination. Actin-Cre-ER mice25 were mated with floxed Sirpa mice. Tamoxifen (100 g to 1 mg) was injected at P0, P15 or P30 to induce the Cre recombinasemediated excision of the Sirpa gene. CD47 knockout mice29 were from W. Frazier (Washington University). Mice used were C57/BL6 background. Both male and female mice were used in our experiments. We did not detect any significant differences between males and females. All animal care and use was in accordance with the institutional guidelines and approved by the University of Michigan Committee on Use and Care of Animals. Immunohistochemistry. Cultures were fixed with methanol for 3 min at 20 C or with 14% paraformaldehyde (PFA) for 10 min at 37 C and stained as described17. Mouse brains were fixed for 24 h with 4% PFA in PBS. Sagittal sections of 20 m thickness were cut in a cryostat and stained. For immunostaining for PSD95, mouse brains were fresh-frozen and sectioned. Sections were then fixed with methanol for 5 min at 20 C and stained. Dilutions and sources of antibodies were as follows: anti-VGLUT1 (1:5,000; Millipore; AB5905)17,26, anti-PSD95 (1:250; NeuroMab; 75-028)17, anti-VGAT (1:1,500; Synaptic Systems; 131003)17, anti-MAP2 (1:3,000; Sigma-Aldrich; M4403)17, anti-neurofilament (1:1,000; Covance; SMI-312)17, anti-bassoon (1:500; Enzo Life Sciences; ADIVAM-PS003)17,26, anti-GFP (1:1,000; Millipore; AB16901)17,26,41, anti-GFAP (1:500; Synaptic Systems; 173002)17, anti-NeuN (1:500; Millipore; MAB377)17, anti-calbindin (1:500; Sigma; C9848)17, anti-synapsin (1:2,000; a kind gift from P. Greengard and A. Nairn, Rockefeller University)15,16, antibody Py (1:50; a kind gift from M. Webb and P.L. Woodhams)17,45, anti-CD47 (1:200; BD; miap301)16,35, polyclonal anti-SIRP (against the SIRP C-terminal domain; 1:200; Upstate Biotechnology; 06-729)16,34 and monoclonal anti-SIRP (clone p84; 1:200; BD)16,46. Clone p84, which recognizes the extracellular domain of SIRP and stains cell surface SIRP, inhibits CD47-SIRP interactions46, suggesting that the epitope of this antibody may be close to the CD47 binding site of SIRP, which is its most distal immunoglobulin domain36. Imaging and quantification. Twelve-bit images at a resolution of 1,376 1,032 pixels were acquired on an Olympus BX61 epifluorescence microscope using 4 (2,226 1,670 m images), 10 (890 668 m), 20 (444 333 m) and 40 (221 166 m) objective lenses and an F-View II CCD Camera (Soft Imaging System). Alternatively, twelve-bit images at a 1,024 1,024 pixel resolution were acquired on a confocal microscope (Olympus, FV1000) using 40 objective lens with zoom 1.5 (211.8 211.8 m). All images for each experiment were acquired with identical exposure time and detector gain. For images of hippocampal sections stained for synaptic proteins, the average signal intensities of staining in the stratum lucidum and stratum radiatum were calculated with MetaMorph software. The average signal intensity in the

fimbria of the hippocampus was calculated and subtracted as the background. The intensity of the background was not obviously different between Sirpa/ and control mice. For images of cultured neurons stained for synaptic proteins, the staining intensity of the dendritic shaft was calculated and subtracted as the background. Size and intensity of puncta were quantified using the MetaMorph or ImageJ software17. Electron microscopy. P15 SIRP conditional knockout mice and littermate controls were injected with tamoxifen. At P29, the mice were perfused transcardially with Karnovskys fixative17. Hippocampi were removed and 1-mm cubes from the stratum radiatum of the CA3 region were dissected and pro cessed for electron microscopy. Thin sections (70 nm) were cut and imaged with a Philips CM100 electron microscope at 60 kV. Digital images were captured with a Hamamatsu ORCA-HR digital camera system operated with Advanced Microscopy Techniques Corp. software. Plasmid constructs and recombinant proteins. A mouse Sirpa cDNA (IMAGE: 5368250) was obtained from ATCC and cloned into the APtag5 expression vector (GenHunter) as described previously16. A cDNA encoding the mutant (sheddingresistant) Sirpa (ref. 47) was generated by PCR using the following primers: 5-GGATATCGATTACAAGGACGACGATGACAAGACCCACAACTGGA ATGTCTTCATCG-3 and 5-AGGTATCGATATCCCCTTGATCACTCGAGT GG-3. This replaced the juxtamembrane amino acids SMQTFPGNNA in the SIRP protein with the Flag epitope amino acids DIDYKDDDDK. The expression plasmids for FGF22 and synaptophysin-YFP were described previously15,17. The expression plasmid for Neuroligin1 was from G. Rudenko (University of Michigan), that for CD47 (IMAGE: 4187965) was from Open Biosystems, and that for Cre was from D. Goldman (University of Michigan). Soluble SIRP proteins (sSIRP) were produced by transiently transfecting SIRP extracellular-domain plasmids (Ext-SIRP) into HEK cells and purifying secreted SIRP proteins from culture media as described16. Recombinant FGF22 was from R&D Systems. Immunoprecipitation. Hippocampal neurons were cultured for 1012 d, after which the medium was replaced by new medium (control) or medium containing 50 mM KCl, 50 M bicuculline, 1 M TTX, 5 M KN62, 5 M KN93, 10 M nifedipine, 10 M GM6001 and/or TIMPs (0.5 g/ml each of TIMP1 and TIMP2). After incubation for 13 d with these treatments, media and cells were collected. The media were precleared with Immobilized Protein-L (Pierce) and incubated with 1 g in 2 l of anti-SIRP extracellular domain antibody (p84)16,35,47 for 4 h at 4 C. The immune complexes were precipitated with Protein-L and the immunoprecipitates were subjected to western blotting as described below to assess the amount of secreted SIRP. Western blotting. Cells, collected as described above, were lysed on ice for 1 h in 200 l of lysis buffer (1% Nonidet P-40, 50 mM Tris buffer, pH 8.0) with a protease inhibitor cocktail tablet (Roche). Dissected hippocampi were lysed by homogenization in 10 volumes of lysis buffer per gram of tissue. Lysis buffer was 1% Triton, 50 mM Tris buffer (pH 7.4) and 150 mM NaCl with a protease inhibitor cocktail tablet. The immunoprecipitates and lysates were subjected to SDS-PAGE. Equal amounts of lysate from each group were applied to the gel, as confirmed by assaying for -tubulin. Proteins were transferred to a polyvinylidene fluoride (PVDF) membrane and probed using anti-SIRP extracellular domain antibody (p84; 1:200; BD)16,35,47 and anti--tubulin antibody (1:5,000; Sigma; T6074)16,17,42. The proteins were visualized by chemiluminescence (GE Healthcare) and the band intensities were quantified with ImageJ software. Synaptic protein fractionation. The fractionation protocol was adapted from a previous report48. Cortex (300350 mg) from P21 mice was homogenized in 1.5ml of homogenization buffer (0.32 M sucrose, 1 mM NaHCO3, 1 mM MgCl2, 0.5 mM CaCl2, protease inhibitor). Homogenate was then adjusted to 1.25 M sucrose and 0.1 mM CaCl2 in a total of 5 ml. Homogenate was overlaid on 5 ml of 1 M sucrose and spun at 100,000g (SW41-Ti rotor, Beckman) for 3 h at 4 C. Interface was collected and designated as synaptic membrane fraction (SPM). SPM (500 l) was then added to 2 ml of 0.1 mM CaCl2 and 2.5 ml of 40mM Tris, pH 6, with 2% Triton X-100, and placed on rocking platform for 20min at 4 C. The sample was then spun at 35,000g (SS-34 rotor, Sorvall) for 20 min

2013 Nature America, Inc. All rights reserved.

NATURE NEUROSCIENCE

doi:10.1038/nn.3516

at 4 C and supernatant was collected as the extrajunctional fraction. The pellet was air-dried and resuspended in 1 ml of 0.1 mM CaCl2 and 1 ml of 40 mM Tris, pH 8, with 2% Triton X-100, and placed on rocking platform for 60 min. Resuspended pellets were then spun at 140,000g (SW41-Ti rotor, Beckman) for 30min at 4 C, and supernatant was collected as presynaptic fraction. The insoluble fraction was resuspended in 1 ml 20 mM Tris pH 7.4 with 1% SDS and designated as the postsynaptic fraction. Extrajunctional and presynaptic fractions were acetone-precipitated and resuspended in 1 ml of 20 mM Tris, pH 7.4, with 1% SDS. Synaptic membrane fraction and equivalent volumes of extrajunctional, presynaptic and postsynaptic membrane fractions were then transferred to PVDF membrane and probed with anti-PSD95 antibody (1:500; NeuroMab; 75-028)17, anti-synaptotagmin antibody (1:100; Hybridoma Bank; mab48)49 and polyclonal anti-SIRP antibody (1:500; Upstate; 06-729)16,34. Whole-cell patch-clamp recordings in cultures. Neurons were bathed in HEPES-buffered saline, consisting of 119 mM NaCl, 5 mM KCl, 2 mM CaCl2, 2mM MgCl2, 30 mM glucose and 10 mM HEPES (pH 7.4), supplemented with 1 M TTX and 50 M picrotoxin to isolate mEPSCs. Whole-cell internal solution consisted of 100 mM gluconic acid, 0.2 mM EGTA, 5 mM MgCl2, 2 mM ATP, 0.3 mM GTP and 40 mM HEPES (pH 7.2). Recording pipettes had a resistance of 46 M. Recordings were made with an Axopatch 200B amplifier and collected with Clampex 8.0 (Molecular Devices). mEPSCs were analyzed using Minianalysis 6.0 (Synaptosoft). Acute hippocampal slice preparation and field electrophysiology. Mice were decapitated and the hippocampi were isolated. Transverse slices (400 m) were cut using a tissue chopper (Stoelting) and incubated at 25 C in a humidified chamber for at least 2 h before recording. Slices were then transferred to a recording chamber, maintained at 2728 C and continuously perfused at rate of 1.5ml/min with oxygenated artificial cerebral spinal fluid (aCSF). aCSF consisted of 119mM NaCl, 2.5 mM KCl, 1 mM NaH2PO4, 26.3 mM NaHCO3, 11 mM glucose, 1.3 mM MgSO4 and 2.5 mM CaCl2. In most experiments (except those in Fig. 6), aCSF also contained 50 M picrotoxin during recording. Picrotoxin was perfused for at least 15 min before data were collected. Recording electrodes were pulled from borosilicate capillary glass (1.7 mm o.d.; VWR International), filled with 3 M NaCl and placed in the stratum radiatum of CA1. fEPSPs were stimulated using cluster electrodes (FHC) also placed in the stratum radiatum of CA1. Current was delivered with an ISO-flex stimulus isolation unit (AMPI). Recordings were made with a MultiClamp 700B amplifier, collected and analyzed using Clampfit 10.2 (Molecular Devices). An input-output curve was obtained for each slice by

increasing the stimulus intensity from 0.02 to 0.25 mA. For paired-pulse experiments, the intensity was set at 0.2 mA, which was the maximum response size. To obtain the paired-pulse ratio, two pulses were delivered with an inter-pulse interval from 25 to 200 ms. For LTP experiments, the stimulus intensity was set so that the response size was 50% of maximum, and test stimuli were delivered every 30 s. After 20 min of stable baseline fEPSPs, LTP was induced by delivering four trains (each 1 s in duration) of 100 Hz separated by 30 s each. Statistical analysis. The statistical tests performed were two-tailed Students t-test or one- or two-way ANOVA, as indicated in the figure legend. In the case of a two-way ANOVA, post hoc analysis was done with Tukeys test. All data are expressed as mean s.e.m. No statistical methods were used to pre-determine sample sizes, but our sample sizes were similar to those reported in previous publications in the field1517,41,42. Data distribution was assumed to be normal. From the experiment presented in Figure 6, 3 out of 13 (control) and 2 out of 22 (Sirpa/) data points were excluded because the fiber volley amplitude has changed by more than 10%. No data points were excluded from any other experiments. All steps of the experiments were randomized to minimize the effects of confounding variables. This includes how mice were chosen for injections, order of cell culture treatments, etc. Electrophysiology experiments were done blind. Imaging was done in similar fashions among conditions: fields from brain sections were chosen randomly from the region of interest, and images of cell cultures were taken randomly from all areas of the culture.
43. Schaeren-Wiemers, N. & Gern-Moser, A. A single protocol to detect transcripts of various types and expression levels in neural tissue and cultured cells: in situ hybridization using digoxigenin-labelled cRNA probes. Histochemistry 100, 431440 (1993). 44. Uesaka, T. et al. Conditional ablation of GFR1 in postmigratory enteric neurons triggers unconventional neuronal death in the colon and causes a Hirschsprungs disease phenotype. Development 134, 21712181 (2007). 45. Woodhams, P.L., Webb, M., Atkinson, D.J. & Seeley, P.J. A monoclonal antibody, Py, distinguishes different classes of hippocampal neurons. J. Neurosci. 9, 21702181 (1989). 46. Oldenborg, P.A. et al. Role of CD47 as a marker of self on red blood cells. Science 288, 20512054 (2000). 47. Ohnishi, H. et al. Ectodomain shedding of SHPS-1 and its role in regulation of cell migration. J. Biol. Chem. 279, 2787827887 (2004). 48. Hahn, C.G. et al. The post-synaptic density of human postmortem brain tissues: an experimental study paradigm for neuropsychiatric illnesses. PLoS ONE 4, e5251 (2009). 49. Fox, M.A. & Sanes, J.R. Synaptotagmin I and II are present in distinct subsets of central synapses. J. Comp. Neurol. 503, 280296 (2007).

2013 Nature America, Inc. All rights reserved.

doi:10.1038/nn.3516

NATURE NEUROSCIENCE